The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to run npx @agent-native/core@latest and npx skills@latest to install tools and manage the design connection. These commands fetch and execute code from the npm registry.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands including npx, pnpm, curl, and lsof for bridge management, environment probing, and server communication.
[DATA_EXFILTRATION]: The skill transmits local development environment metadata, such as absolute directory paths (rootPath) and application route manifests, to external endpoints via the open-visual-edit action.
[PROMPT_INJECTION]: The skill ingests untrusted data from the running local application's DOM and route state, which could contain malicious instructions.
Ingestion points: Data enters the agent context from the running local application via the bridge and open-visual-edit action.
Boundary markers: The skill defines a 'hard write boundary' for [Reprompt selection] sequences to restrict unauthorized modifications.
Capability inventory: The skill possesses capabilities for shell command execution, file system access (read/write), and network requests.
Sanitization: The system explicitly excludes sensitive paths like .env and .git from indexing and requires mandatory human-in-the-loop approval for all file write operations.

visual-edit